Controlling striatal function via anterior frontal cortex stimulation

Motivational, cognitive and action goals are processed by distinct, topographically organized, corticostriatal circuits. We aimed to test whether processing in the striatum is under causal control by cortical regions in the human brain by investigating the effects of offline transcranial magnetic stimulation (TMS) over distinct frontal regions associated with motivational, cognitive and action goal processing. Using a three-session counterbalanced within-subject crossover design, continuous theta burst stimulation was applied over the anterior prefrontal cortex (aPFC), dorsolateral prefrontal cortex, or premotor cortex, immediately after which participants (N = 27) performed a paradigm assessing reward anticipation (motivation), task (cognitive) switching, and response (action) switching. Using task-related functional magnetic resonance imaging (fMRI), we assessed the effects of stimulation on processing in distinct regions of the striatum. To account for non-specific effects, each session consisted of a baseline (no-TMS) and a stimulation (post-TMS) fMRI run. Stimulation of the aPFC tended to decrease reward-related processing in the caudate nucleus, while stimulation of the other sites was unsuccessful. A follow-up analysis revealed that aPFC stimulation also decreased processing in the putamen as a function of the interaction between all factors (reward, cognition and action), suggesting stimulation modulated the transfer of motivational information to cortico-striatal circuitry associated with action control.

and at the start of each experimental session, participants completed a second practice block that was exactly the same as the actual paradigm described in the legend of figure 2, only shorter (i.e. 24 trials). Finally, a third block (32 trials) without reward or feedback was administered in the scanner immediately before the actual paradigm started, and during the intake session (see main text: intake session). This third block was used to determine each individual's response window. We calculated the average response times on four trial types (arrow, word x task-switch, task-repeat), during the third practice block. These response times were set as the response deadline during the subsequent run. This was done to account for inter-individual and inter-run differences in response speed and subsequent task difficulty.

Selection and targeting of stimulation sites
To determine the TMS coil positioning for each individual and each brain region, each participant's structural scan was coregistered to the standard SPM8 T1 template (Montréal Neurological Institute; MNI) and segmented using a unified segmentation procedure 27 . This procedure resulted in a set of inverse parameters allowing the conversion of the stimulation targets in group mean MNI coordinates into individuals' native anatomical space. Next, the MNI coordinates for each cortical stimulation site were projected onto each individual's structural scan using a frameless stereotactic neuronavigation system (Localite, Sankt Augustin, Germany). The TMS coil was then placed on the scalp overlying the target coordinates (aPFC, dlPFC, PMC) using the Localite software.
During the intake session, participants were familiarized with the sensation of cTBS over each of these regions. Thirty-nine participants started the intake session. Any participant who reported -or showed -signs of discomfort during this part of the intake session was excluded from further participation. This resulted in the exclusion of eight participants: six participants due to discomfort during stimulation over the aPFC and two due to a more general feeling of discomfort during this part of the intake. As a result, 31 participants started the main experiment (see participants).

Continuous Theta Burst Stimulation (cTBS) protocol
During the determination of the active motor threshold (aMT), participants rested their right hand on a pillow while squeezing a small roll of tape with a pincer grip at 20% of their maximum strength, contracting their first dorsal interosseous FDI muscle continuously. The aMT was defined as the lowest stimulation intensity over the contralateral motor cortex that elicited reproducible MEPs (in at least 5 out of 10 successive stimulations). The aMT was 24%-37% (mean 30.44%, SD 3.61) of the maximum stimulator output.

Preprocessing of task-related fMRI data
Prior to standard preprocessing, realignment was performed using the estimated head motion parameters (least-squares approach, 6 parameters) for the images with the shortest echo, which were applied to echo images for each excitation. The images of all sessions were aligned to the shortest echo of each session, and to the first session. After spatial realignment, the four echo images were combined using echo summation. The combined images were slice-time corrected to the middle slice and segmented using a unified segmentation procedure 27 . The bias corrected T1 image was coregistered to the mean functional image and the transformation matrix from the segmentation procedure was used for normalization to a standard template (MNI). Normalized images were smoothed using an 8 mm full-width half maximum kernel. A study-specific T1 template was generated from an average of all co-registered and normalized T1 images to display the results, using MRIcron software.

First-level analysis
We automatically included any participant with <3mm (one voxel size) of head motion in either direction (N=22). For the 5 participants who moved more than 3mm (but never more than 2 voxel sizes), we made sure movement was gradual, which is easier to correct for. Two participants did show a peak of excessive movement in one of their 6 fMRI runs. We attempted to account for residual head motion by including -for all participants -the six original head motion parameters (3 translation, 3 rotation), their first derivative and the square of the original and first derivative in the model, resulting in 24 motion nuisance regressors 30 . Before including a participant with more than 3mm of head motion, we assessed the activation maps of the affected fMRI run to make sure there were no residual motion artifacts in the individual activation maps. Finally, we repeated the analysis of our primary result to confirm that the 4-way interaction of aPFCBASE-STIM x Reward HIGH vs. LOW x Task SWITCH vs. REPEAT x Response SWITCH vs. REPEAT in the left putamen remained significant after the exclusion of the two participants with excessive head motion.
In addition, we used the mean signal from the white matter and CSF to account for movement-related intensity changes 31 . Finally, a high-pass filter (128s) was used to remove low-frequency signals (e.g. scanner drifts) and an AR(1) model was applied to adjust for serial correlations in the data. Microtime onsets were adjusted to account for the earlier mentioned slice time correction.
Supplementary figure S1 | Whole-brain maps for the effect of aPFC stimulation on the main effect of Reward (in red) and the interaction between Reward, Task switching and Response switching (blue).
Whole-brain maps across the axial (dorsal to ventral) and coronal (anterior to posterior) plane are shown for the effects in figure 4. The effect of aPFC stimulation on the main effect of Reward is shown in red. The effect of aPFC stimulation on the interaction between Reward, Task switching and Response switching in blue. Note that the results are displayed at a low threshold (PUNC < 0.001, t > 3.14), but that statistical significance (FWE-corrected) of the results was assessed in two anatomically defined regions of interest (i.e. restricted to the grey matter of the bilateral caudate nucleus and bilateral putamen).

Functional specificity
To assess functional specificity, we tested whether TMS over the aPFC altered processing in the caudate nucleus exclusively as a function of Reward processing. In other words, we anticipated that TMS over the aPFC would not alter processing in the caudate nucleus as a function of Task switching or Response switching.
In addition, we assessed whether TMS over the aPFC altered processing in the putamen exclusively as a function of the interaction between Reward, Task switching and Response switching, and not as a function of any other effects (e.g. the effect of Reward, Task switching, or Response switching, or the interaction between Reward and Task switching).
The effects of aPFC stimulation on reward-related processing in the caudate nucleus and the effects of aPFC stimulation on the interaction between Reward, Task switching and Response switching in the putamen were functionally specific: aPFC stimulation did not decrease striatal processing as a function of Task switching (caudate nucleus: k = 0, putamen: PSVC_-FWE = 0.165, k = 2) or Response switching (caudate nucleus: PSVC_-FWE = 0.150 k = 3, putamen: k = 0). In addition, the effects of aPFC stimulation did not decrease striatal processing as a function of the interaction between Reward and Task switching.

Anatomical specificity at the level of the striatum
We assessed quantitatively the anatomical specificity of an effect in a region. For example, we aimed to assess whether the effect of aPFC stimulation on the effect in the left putamen (figure 4b) was different from the same effect in the left caudate nucleus. To avoid double dipping 32 , we derived beta values from an independent anatomical ROI (i.e. independent from the activated cluster). Values were entered into a repeated measures GLM in SPSS with the factors stimulation (aPFCSTIM vs. BASE), ROI (caudate nucleus vs. putamen) and either 1) Reward (HIGH VS. LOW) (time-locked to the reward cue) or 2) Reward (HIGH VS. LOW), Task (SWITCH VS. REPEAT) and Response (SWITCH VS. REPEAT) (all time-locked to the target).
A direct comparison between the effect of Reward in the right caudate nucleus and the effect of reward in the right putamen revealed that the (trending) effect in the right caudate nucleus was specific to the caudate nucleus: ROI (right caudate nucleus vs. right putamen) x Stimulation (aPFCBASE-STIM) x Reward: F(1,26) = 4.937, The effect in the putamen during the integration of Reward, Task and Response was anatomically specific: these results are discussed in the main text.

Anatomical specificity at the level of the cortex
To test whether the effect of aPFC stimulation on Reward-processing in the caudate nucleus was anatomically specific at the level of the cortex, we submitted the data from each session for the factor Reward to a GLM with the additional factor Site (aPFC, dlPFC, PMC). This allowed us to assess whether the (marginal) effect of aPFC stimulation (BASE-STIM) on reward-related processing in the caudate nucleus (figure 4a) was different during the aPFC session compared to the other two session. This interaction test (Stimulation x Site (i.e. aPFC BASE -STIM > dlPFC BASE -STIM = PMC BASE -STIM) x Reward) revealed no significant voxels in the whole-brain or the after applying a SVC in the striatum.
To test whether the effect of aPFC stimulation on the interaction between Reward, Task and Response was significantly different compared to the data from the dlPFC (BASE -STIM) and PMC (BASE -STIM) sessions, we submitted the data from each session for the Reward x Task x Response contrast to a GLM with the additional factor Site (aPFC, dlPFC, PMC), resulting in the following interaction test: Stimulation x Site (i.e. aPFC BASE -STIM > dlPFC BASE -STIM = PMC BASE -STIM) x Reward x Task x Response. This analysis revealed one significant cluster in the left putamen: PSVC-FWE = 0.0248, t = 3.92, z = 3.83, peak x, y, z = -26, -8, 12). The whole-brain map for this interaction is shown in figure S3a at a threshold of PUNCORRECTED < 0.001. To visualize this interaction we extracted the beta values as described elsewhere (see order effects). For an unbiased representation of the data, we extracted the beta values from the left anatomically defined putamen. We plotted the interaction between Reward, Task switching and Response switching on the Y-axis of figure S3b, separately for each stimulation site (aPFC, dlPFC, PMC) and for the stimulation fMRI vs. baseline fMRI run.

Order effects: Methods
We aimed to assess whether the results in the caudate nucleus (main effect of reward) and putamen (the Reward x Task switching x Response switching interaction), were dependent on the order in which participants performed the baseline and stimulation fMRI run (Stimulation order), i.e. stimulation fMRI followed by a baseline fMRI (N=13) or the opposite arrangement (N=14) (figure 1). We reasoned that any residual effect of the inhibitory TMS protocol in those who performed the stimulation fMRI run first, would be evident in a reduced effect during the baseline fMRI run in this group of participants. In addition, we aimed to assess whether we could find any evidence to suggest that the order of the stimulation Site (Site order), i.e. on which day the aPFC stimulation took place (i.e. during the 1 st , 2 nd or 3 rd session), had an effect on the Reward effect or on the interaction between Reward, Task switching, and Response switching.
We extracted the beta values during the Reward cue (high and low Reward) from the cluster in the right caudate nucleus as presented in figure 4a. Next, using SPPS software (IBM SPSS Statistics 23), we performed a repeated measures GLM with the within-subject factors Stimulation (aPFC stimulation vs. aPFC baseline), and Reward (high vs. low) and the between subject factors Stimulation order and Site order (as described above).
From the cluster in the left putamen, as presented in figure 4a, we extracted the beta-values during target for the interaction between Reward, Task switching and Response switching and entered these variables as well as Stimulation (aPFC stimulation vs. aPFC baseline) as within-subject factors. We added the between subject factors Stimulation order and Site order (as described above).
In addition, because we observed a Site (aPFC, dlPFC, PMC) x Stimulation (stimulation vs. baseline) x Reward x Task x Response interaction in the putamen (figure S3a), we also assessed whether that interaction varied either as a function of Stimulation order or Site order. Because the voxels from which the beta values were extracted were based on the interaction between aPFC BASE-STIM, we assessed these effects in an unbiased region: the anatomically defined left putamen.

The effect of aPFC stimulation on Reward processing in the caudate nucleus
We did not find any evidence to suggest that the effect of aPFC stimulation vs. aPFC baseline on the (trending) Reward-related signal in the caudate nucleus (figure 4a) was different in those participants who started the session with the baseline fMRI run (figure S2a dark red bars: baseline followed by stimulation) compared with those who started the session with cTBS followed by the stimulation fMRI run (figure S2a light red bars: simulation followed by baseline). More specifically, the Reward x Stimulation (aPFC BASE-STIM) x Baseline order interaction was not significant (F(1,21)<1. In addition, there was no evidence that the effect of aPFC stimulation vs. aPFC baseline on the Reward-related signal in the caudate nucleus was dependent on the session number (1 st , 2 nd or 3 rd ) in which the aPFC was stimulated (F(2,21) = 2.351, p >0.05. x Reward x Task x Response x Baseline order interaction was not significant (F(1,21) = 3.672, p > 0.05).

Supplementary figure S2 | Effects of order (baseline vs. stimulation fMRI run)
Plots of beta-values extracted from a) the right caudate nucleus cluster (see figure 4a) for the contrast high vs. low Reward and for b) the left putamen cluster (see figure 4b) for the contrast high vs. low Reward x Task switching (switch vs. repeat) x Response switching (switch vs. repeat). Results are shown separately for participants who performed the baseline fMRI run prior to the stimulation fMRI run (light bars) and those who performed the stimulation fMRI run before the baseline fMRI run (dark bars).
Moreover, the observed (not significant) pattern was opposite to what we would expect if the cTBS stimulation had carried over to the baseline fMRI run. If the 90 minutes between cTBS stimulation and the cTBS baseline fMRI run had not been sufficient, we would expect to see the opposite pattern from the one observed here: I.e., we would expect a smaller difference between the aPFC stimulation and the subsequent aPFC baseline session in those participants who started with the stimulation run (the lighter bars in figure S2b).
In addition, there was no evidence that the effect of aPFC stimulation vs. aPFC baseline on the Reward, Task, Response-related signal in the putamen (figure 4b) was dependent on the session number (1 st , 2 nd or 3 rd ) in which the aPFC was stimulated (Stimulation (aPFC BASE-STIM x Reward x Task x Response x Site order: F(2, 21) <1).
Finally, we repeated the analyses and included data from all sites (aPFC, dlPFC and PMC), to assess whether we could find evidence that the Site x Stimulation x Reward x Task x Response interaction in the putamen ( figure   S3) was modulated by either Stimulation order or Site order. We did not find evidence for these effects (both F's < 1).

Inspection of figure S2b suggests an effect in the putamen as a function of the interaction between
Reward, Task switching and Response switching during the baseline fMRI runs of the aPFC and PMC session, but not during the dlPFC session. Indeed, a direct contrast between the baseline aPFC and the baseline dlPFC session (grey bars) revealed that the BOLD signal in the left putamen was lower during the dlPFC baseline session (Site (aPFC BASE vs. dlPFC BASE) x Reward x Task x Response: PFWE-SVC = 0.019), but that there was no evidence of a difference between the baseline runs during the aPFC and PMC sessions (Site (aPFC BASE vs. PMC BASE) x Reward x Task x Response: PFWE-SVC > 0.05). To assess directly in the left putamen whether the effect in figure S2 can be explained by a difference in BOLD response during the baseline sessions, we performed a direct test of the Reward x Task x Response x Stimulation interaction between the aPFC and PMC session (i.e. sessions with comparable baselines). This revealed a significant cluster in the left anatomical putamen (aPFC BASE-STIM > PMC BASE-STIM x Reward x Task x Response: PFWE-SVC = 0. 026, k = 26, t = 3.63, z = 3.55, x, y, z peak = -26, -8, 12). This analysis confirms that the significant 4-way interaction, reported in the main text, i.e. aPFC BASE -STIM > dlPFC BASE -STIM = PMC BASE -STIM) x Reward x Task x Response, was not driven by the dlPFC session.

Supplementary figure S3 | Effect of aPFC vs. dlPFC and PMC stimulation (vs. baseline) for the interaction between Reward, Task switching and Response switching a)
Brain maps for the interaction effect of cTBS (stimulation vs. baseline) x Site (aPFC, dlPFC, PMC) x Reward x Task switching x Response switching (shown at PUNC < 0.001, note that the cluster in the left putamen is significant at PFWE-SVC = 0.026). Color scales reflect t-values. b) Plots of the beta-values extracted from the anatomically defined left putamen for the same contrast as in figure S3a.

Differences in the baseline run do not account for the effect of aPFC (vs. dlPFC and PMC) stimulation in the putamen
When inspecting the two sessions with equal baselines in figure S3b, it can be seen that the effect of aPFC BASE-STIM vs. the PMC BASE-STIM is driven by a combination of a numerical increase in the Reward x Task x Response -related BOLD response during the aPFC baseline fMRI run and a numerical decrease in the Reward x Task x Response -related BOLD response during the aPFC stimulation fMRI run. This was confirmed by a post-hoc test: direct comparison of the aPFC stimulation vs. PMC stimulation runs or the aPFC baseline vs. PMC baseline runs did not result in a significant cluster in the putamen. Combined, these effects resulted in a significant effect. In the absence of a baseline session, we would not have been able to detect this effect..
Thus, the inclusion of the baseline fMRI runs in our design has enabled us to take into account variation in task-related BOLD response within the same individuals across different days. This enabled us to increase the sensitivity of our measurements, quantifying inter-session variance that would otherwise have been undetectable.

Methods: Statistical analysis of behavioral data
The first trial of each block, trials with extremely fast responses (<100ms), and trials on which participants failed to respond were excluded from analysis. In addition, trials on which the response was incorrect were excluded from RT analyses. The RT and error rate data violated a normal distribution, which was not resolved after log and arcsine transformations, respectively (Shapiro-Wilk p < 0.05). Therefore, effects that reached significance were submitted to a non-parametric Wilcoxon signed-rank test to assure that the repeated measures GLM did not reflect false-positives. We did not observe any false-positives.

Discussion of behavioral data
A number of studies have reported that task-switching and response switching are not independent 1,2 . This is substantiated by our behavioral data. The task-switch cost observed was larger on response-repeat trials, in agreement with previous reports 1,2 . In addition, we observed a behavioral response switch benefit: participants were faster and more accurate when responding on a response switch compared with a response repeat trial (figure S4). When inspecting figure S4, it becomes clear that there was a considerable behavioral cost associated with trials on which the task switched, but the response remained the same. It appears that this increased error rate on response repeat/task switch trials is driving both the interaction and the main effect.